2.1 Sustainability assurance versus carbon assurance

The increasing demand from various stakeholders on sustainability-related issues has pressured firms around the world to publish their sustainability reports. Simnett et al. (2009) refer to such reports as relevant non-financial information with the following six categories in the Global Reporting Initiative (GRI) 2007: economic, environment, human rights, product responsibility, labour, and society. However, given the public concerns over the credibility of such reports, the authors find that firms also voluntarily adopt sustainability assurance to enhance their credibility. There are multiple types of sustainability assurance that correspond to these categories of sustainability information (Hazaea et al., 2022), such as social assurance (Gao & Zhang, 2006), environmental assurance (Zhang et al., 2022), and carbon assurance (Bui et al., 2021; Luo et al., 2023).

While environmental assurance includes various components such as energy, water usage, and carbon emissions (Zhang et al., 2022), carbon assurance only focuses on carbon emissions assurance. As carbon reporting and assurance is governed by particular accounting standards (e.g., International Standard on Assurance Engagements-ISAE 3410, SEC Guidance Regarding Disclosure Related to Climate Change), which require specific knowledge and expertise, there is a shift to examine carbon disclosure and assurance specifically (Luo et al., 2023). However, like other types of sustainability disclosure, it is challenging to assess the reliability and credibility of carbon reporting (Freedman & Jaggi, 2005). Furthermore, such information asymmetry in carbon reporting cannot be resolved with financial auditing, thus firms are motivated to adopt carbon assurance (Fan et al., 2021).

The literature on carbon assurance is still limited and mainly focuses on carbon assurance practice and its determinants (He et al., 2022a). Particularly, as carbon assurance is voluntary, firms can choose to adopt different carbon assurance levels (e.g., Scope 1, Scope 2) and use various assurance providers (e.g., audit firms, consulting firms) (Datt et al., 2018; He et al., 2022a). Such assurance practices are also influenced by various factors. For example, Datt et al. (2018) analyse a global sample of firms across 44 countries and find that firms that implement carbon reduction initiatives or have higher carbon disclosure scores tend to adopt carbon assurance. Similarly, Simic et al. (2023) show that sustainability-related compensation or higher director compensation is associated with voluntary carbon assurance in UK firms. Tang (2019) examines Chinese firms and documents that governmental green funding and the forming of carbon institutions are associated with carbon assurance. In summary, while prior studies have offered important insights into carbon assurance and its determinants, little is known about how carbon assurance impacts various firm outcomes, especially credit ratings.

2.2 Carbon assurance and credit ratings

According to signalling theory (Connelly et al., 2011; Spence, 1978), when information asymmetry is present, high-quality individuals or organisations tend to send “costly” and “observable” information to reduce the asymmetric information problem among parties and signal their high quality. In the context of environmental reporting, firms and executives know best about their carbon activities. Despite the carbon disclosures, such disclosures can be difficult for external stakeholders to verify (Fan et al., 2021), leading to information asymmetry between firms and intended users. Fan et al. (2021) document that firms with high carbon information asymmetry voluntarily adopt external assurance for their GHG statements to alleviate the asymmetry problem. This adoption of assurance a signals the firms' strong commitment to sustainability (Clarkson et al., 2019).

A quality signal must possess two characteristics: First, it must be costly (i.e., the signal must be costly for signallers to produce to avoid copying or cheating) and second, it must be observable (i.e., the signal must be observable by receivers) (Connelly et al., 2011; Spence, 1978). The assurance process is costly due to the fees paid to the auditor or specialist and the efforts devoted by the management team during the process (Baier et al., 2022; Kausar et al., 2016). It is also observable, as the assurance statement is available as part of the company reports. As a result, carbon assurance can act as a strategic signal for high-quality firms (in terms of carbon performance). These high-quality firms tend to have lower regulatory and climate-change risks related to their carbon emissions, which can improve their credit ratings.

The current carbon reporting framework and relevant assurance requirements are challenging for intended users to verify and are also subject to criticism (Fan et al., 2021; Kaplan & Ramanna, 2021); firms can take advantage of this to send false signals to users. For example, Kaplan and Ramanna (2021) claim that the current carbon emission reporting of some firms ignores emissions from their supply/distribution chains. Moreover, although the assurance statement is available it should be noted that the outcomes may vary depending on how recipients perceive and interpret the assurance signal and its credibility. (Baier et al., 2022). “Not surprisingly, auditors of these reports often resort to double negatives ‘We found no evidence of misreporting in the company's ESG report’ and the reports themselves have had little impact on either corporate actions or external stakeholders” (Kaplan & Ramanna, 2021, p. 122). Consequently, low-quality firms can take advantage of these inconsistencies and choose to disclose selective information and adopt external assurance to “greenwash” their actual carbon activities and show their legitimacy (legitimacy theory) (Clarkson et al., 2019; Kaplan & Ramanna, 2021). In such cases, carbon assurance can be associated with more risks due to potential greenwash problems, reducing firm credit ratings. Given the two competing views on the relationship between carbon assurance and credit ratings, we posit the following hypothesis:

3.1 Data and sample

The sample for the study covers US publicly traded firms over the period 2007–2017 due to COMPUSTAT credit rating data only being available until 2017. The data is collected from a number of different sources. Data on carbon emissions is collected from the CDP database, a comprehensive database that has been used extensively in carbon studies (Bui et al., 2021). Data on firm credit rating is collected from the COMPUSTAT database.

The UK-based non-profit organisation, CDP, partners with institutional investors and supply chain members to prompt large corporations to disclose carbon emissions information. In 2002, CDP sent letters co-signed by institutional investors to CEOs of all Financial Times Global 500 companies, urging completion of a questionnaire covering: (i) climate change risks and opportunities; (ii) the firm's climate change strategy; and (iii) greenhouse gas emissions. The survey became an annual practice after follow-up questionnaires in 2003 and 2005.

In 2017, disclosures were received by CDP, collectively representing more than 55% of firms based on global market capitalisation. Our initial sample consists of 7579 firm-year observations. First, we drop 4214 firm-year observations that do not have any identifier (e.g., ISIN). Subsequently, 482 firm-year observations are dropped due to missing independent variables. Finally, 346 firm-year observations are dropped due to missing dependent and control variables. Our ultimate sample comprises 441 firms, encompassing 2537 firm-year observations. Then, for each variable (except carbon assurance), we Winsorise both the top and bottom 1% of observations to reduce the effect of outliers on the results. The sample selection process is described in Table 1 (Panel A). We exhaust the two sources to prepare the dataset necessary to test the study hypothesis.

Table 1 Panels B and C reports the sample distribution by year and industry. 2014, 2015, 2016, and 2017 have above 400 firm-year observations each year. 2012 and 2013 have more than 300 firm-year observations each year. The rest of the sample has less than 300 observations per year. Regarding the sample distribution by industry, the industrial sector reports the highest number of observations (15.32%), followed by information technology (14.71), financials (13.69%), consumer discretionary (12.55%), and consumer staples (11.45%). Conversely, communication services (2.20%), energy (3.39%), and real estate (3.79%) have the fewest observations.

3.2 Measurement of variables

We use Standard & Poor's (S&P) rating as a proxy for firm credit rating (RATING). The rating reflects the overall quality of the firm's outstanding debt, both private and public. We convert the S&P ratings into numerical scores, where 22 indicates an AAA rating (highest rating) and 1 represents a D rating (lowest rating i.e., defaulting on a debt payment). Consistent with previous studies on credit ratings (Ma et al., 2021), we use S&P's long-term issuer credit rating for a firm. Following Bui et al. (2021), we use two proxies to capture carbon assurance: assurance level and assurance percentage. The first proxy of carbon assurance is measured as the assurance level (ASSUR_LEV) for both Scope 1 and Scope 2 emissions, ranging from 0 to 4 for five different levels of carbon assurance (low to high): 0 for no verification, 1 for limited assurance, 2 for moderate assurance, 3 for reasonable assurance, and 4 for high assurance. The ranking scores are then added across the two emission scopes to derive our measure of carbon assurance.

Similarly, the second proxy for carbon assurance is measured as the carbon assurance percentage (ASSUR_PER) for both Scope 1 and Scope 2 emissions (ranging from 0 to 100), reflecting the percentage of Scope 1 and Scope 2 emissions verified in a given year. Particularly, the assurance percentage is computed for each Scope 1 and Scope 2 emissions and then averaged, the total of which represents our carbon assurance percentage measure (Bui et al., 2021).

The choice of control variables is guided by relevant previous studies on credit ratings (for example, Bonsall & Miller, 2017; Ham & Koharki, 2016; Jiraporn et al., 2014; Ma et al., 2021). Firm size (SIZE), computed as the natural logarithm of total assets, is controlled because a firm's credit rating is found to be positively associated with firm size (Ashbaugh-Skaife et al., 2006; Bhojraj & Sengupta, 2003). Profitability (ROA), measured as earnings before interest and tax divided by total assets, is included because highly profitable firms often have higher credit ratings (due to a firm's ability to pay their interest and debt) (Ashbaugh-Skaife et al., 2006). Considering that the coverage of interest and debt is related to firm credit rating owing to the agency cost of debt (Jensen & Meckling, 1976; Myers, 1977), we control for the interest coverage (INTEREST_COVE) and leverage (LEV). We also control for Tobin's q (TOBINSQ) as we expect it to be positively associated with credit rating (Lim & Mali, 2024). Following Elyasiani et al. (2010), we control return volatility (ROA_SD) because high-volatility firms are likely to have a higher default risk (Ma et al., 2021). Since capital expenditure (CAPX) and research and development intensity (R&D_INTENS) have future value implications (Jiraporn et al., 2014; Klapper & Love, 2004), we include CAPX, which is computed as capital expenditure divided by total assets, and R&D_INTENS, measured as research and development expenditure divided by total assets. Finally, firms with a low default risk (high Z-SCORE) and low forecast error (FORECAST_BIAS) tend to have a better credit rating (Jiraporn et al., 2014; Lamy & Thompson, 1988).

Research also shows that credit ratings of firms are associated with carbon emissions (Safiullah et al., 2021) and relevant governance characteristics such as board size (Lin et al., 2020), gender diversity (Kinateder et al., 2021), board independence, and CEO duality/CEO power (Ashbaugh-Skaife et al., 2006; Papadimitri et al., 2020). Particularly, firms with high carbon emissions are found to face more cash flow uncertainty, lowering their credit ratings (Safiullah et al., 2021). On the other hand, firms with strong corporate governance tend to have better credit ratings (Ashbaugh-Skaife et al., 2006). Following that, we further control for carbon emission (CARBON EMISS) (i.e., firm-level total carbon emissions including Scope 1 and Scope 2 carbon emissions) and the relevant corporate governance variables including board size (BSIZE), gender diversity (BGENDERDIV), board independence (BINDEPENDENCE), CEO duality, and whether the current chairman is the previous CEO (CHAIREXCEO).

Following Datt et al. (2020), we include industry dummies (2-digit SIC industry) in our analysis to control for any differences in firms' inherent business risk and the unobserved heterogeneity across firms in various industries. We also include year dummies to control for changes across time in the firms' operational activities. All variable definitions are summarised in the Appendix.

3.3 Research design

3.4 Descriptive statistics and correlation

We present the descriptive statistics and correlation matrix in Tables 2 and 3, respectively. The mean value of credit ratings in our sample is 7.498, with a standard deviation of 7.664. Our carbon assurance level, which has an average (standard deviation) value of 1.585 (1.769) (Table 2), is relatively comparable to the carbon assurance variable in Bui et al. (2021) (i.e., mean of 1.77 and standard deviation of 2.02). The mean value of our carbon assurance percentage is 59.394, with a standard deviation of 47.032. The correlation matrix shows that credit ratings are positively correlated with the two independent variables: carbon assurance level and assurance percentage, and such control variables as firm size, return on assets, leverage, Tobin's q, capital expenditure, Z-score, carbon emissions, board size, and CEO duality, but negatively correlated with return volatility, forecast bias, and board gender diversity (Table 3). The carbon assurance level is positively correlated with carbon assurance percentage, firm size, carbon emissions, board size, board independence, but negatively correlated with interest coverage ratio, return volatility, capital expenditure, research and development expenditure, and Z-score. Similarly, the carbon assurance percentage is positively correlated with firm size, capital expenditure, carbon emissions, board size, board gender diversity, CEO duality, chairman as ex CEO but negatively correlated with Tobin's q, return volatility, research and development expenditure, and forecast bias.

4.1 Baseline regressions

Table 4 reports the results of the regressions of the firm-level credit rating on carbon assurance, controlling for firm-specific characteristics, industry, and year-fixed effects. Panel A presents the contemporaneous results and Panel B shows the 1-year lagged effects (where all independent and control variables are lagged by 1 year). Results from the two panels reveal a significantly positive relationship between carbon assurance and credit ratings. The significant relationship is also robust to two carbon assurance proxies (carbon assurance level in models (1) and (3) and carbon assurance percentage in models (2) and (4)). Particularly, regarding the contemporaneous effects (Panel A, Table 4), the coefficients of carbon assurance level and percentage are positive (0.029, z-statistic = 1.86, and 0.002, z-statistic = 3.77, respectively) and statistically significant, suggesting that firms that have their carbon report and emissions assured tend to receive a higher credit rating in the same year. The relationship is also economically significant, with a one-standard-deviation increase in the carbon assurance level associated with a 5.13% (1.769*0.029) increase in the firm credit rating (model (1), Table 4). Similarly, a one-standard-deviation increase in the carbon assurance percentage is associated with a 0.094% (47.032*0.002) increase in the rating (model (2), Table 4). As shown in Panel B (Table 4), the coefficients of carbon assurance level and percentage in the previous year are also positive (0.031, z-statistic = 1.84 and 0.002, z-statistic = 2.54, respectively) and economically significant. A one-standard-deviation increase in the carbon assurance level is associated with a 5.48% (1.769*0.031) increase in the firm credit rating (model (3), Table 4) and a one-standard-deviation increase in the assurance percentage is associated with a 0.094% (47.032* 0.002) increase in the credit rating (model (4), Table 4). This implies that the carbon assurance practice in the previous year also improves the firm credit rating in the current year. Overall, these findings support signalling theory (Connelly et al., 2011; Spence, 1978), that is, high-quality firms (in terms of environmental performance) tend to have their carbon report/emissions assured (signals) to reduce the information asymmetry problem (Fan et al., 2021) and signal their good quality (Clarkson et al., 2019). The positive relationships also suggest that carbon assurance can be a strategic signal for a firm credit rating, justifying the benefits of adopting carbon assurance and the proposed carbon assurance requirements.9

We also observe significant relationships between the various control variables and firm-level credit ratings. Particularly, firm size, z-score, board gender diversity, and CEO duality are positively associated with credit ratings, while interest coverage, ROA standard deviation, return volatility, forecast bias, and carbon emissions are negatively associated with the ratings. These results are consistent with a number of relevant studies (e.g., Ashbaugh-Skaife et al., 2006; Bhojraj & Sengupta, 2003; Jiraporn et al., 2014). For example, large firms are found to have better credit ratings as they are likely to face lower risks (Ashbaugh-Skaife et al., 2006; Bhojraj & Sengupta, 2003). Firms with a low default risk (high Z-score) tend to have a better credit rating (Jiraporn et al., 2014; Lamy & Thompson, 1988). On the other hand, high-volatility firms are likely to have a higher default risk (Elyasiani et al., 2010). Consistently, we show that volatility is also negatively associated with firm credit ratings.

4.2 Robustness tests

4.3 Channel analysis

We also examine the channel through which carbon assurance influences firm-level credit ratings. One potential channel is that carbon assurance reduces information asymmetry, attracting more analyst following (Cuadrado-Ballesteros et al., 2017; Jiraporn et al., 2012), which in turn can lower default risk and improve credit ratings (Cheng & Subramanyam, 2008). Cuadrado-Ballesteros et al. (2017) find that assured sustainability reports are associated with lower information asymmetry between informed and uninformed stakeholders, as assurance helps improve the credibility and accuracy of such reports. At the same time, a low level of information asymmetry or a high level of transparency can create a good information environment, attracting more analyst coverage (Jiraporn et al., 2012). Financial analysts act as information intermediaries in the market (Healy & Palepu, 2001), thus having more analysts following reduces cash flow uncertainty and information risk, lowering firm default risk and improving credit ratings (Cheng & Subramanyam, 2008). Indeed, prior studies document a positive (negative) relationship between analyst following and credit ratings (default risk) (Cheng & Subramanyam, 2008) or a significant moderating role of analyst coverage in default risks (Baghdadi et al., 2020).

Following Cheng and Subramanyam (2008) and Baghdadi et al. (2020), we measure analyst following by the number of financial analysts issuing an annual forecast for a firm in a fiscal year. This information is available from the Institutional Brokers' Estimate System (IBES) database. We then regress analyst following on carbon assurance and other control variables as in Equations (1) and (2), following prior studies (Baghdadi et al., 2020, 2023). Table 10 (1st stage) shows our channel analysis results, which confirm a significantly positive relationship between analyst following and the two carbon assurance proxies (0.150, z-statistic = 2.14 for carbon assurance level, and 0.029, z-statistics = 2.13 for carbon assurance percentage). Such findings suggest that firms with a high level or percentage of carbon assurance tend to have better analyst coverage. In the second stage, we show that analyst following is positively associated with firms' credit ratings (0.001, z-statistics = 2.52 for carbon assurance percentage), supporting our channel analysis.

4.4 Endogeneity tests

We document a significantly positive relationship between carbon assurance and firm-level credit ratings. Although this finding is robust to various alternative measures and tests, including using lagged variables, the relationship can be subject to endogeneity problems. Particularly, credit ratings and carbon assurance can be jointly correlated with unobserved variables (omitted variables), or firms with high credit ratings tend to have more resources to adopt a higher level of carbon assurance (reverse or simultaneous causality) (Bascle, 2008; Hamilton & Nickerson, 2003; Hill et al., 2020). We employ three methods to address potential endogeneity issues: propensity score matching, instrumental variables, and Heckman selection bias. Table 11 presents the results of our propensity score matching. Following prior studies (Li, 2013; Smith, 2016), we first define (control group) the treated group as firms that (do not) adopt carbon assurance in a given year (firm-year observations). We then estimate the probability that a firm has adopted carbon assurance (propensity scores) using a logit regression (carbon assurance dummy with a value of 1 if a firm adopts carbon assurance in a particular year and 0 otherwise) with all the covariates in our baseline regression model. Next, based on the nearest propensity scores (the nearest neighbour approach), we stratify data to match firms that adopt carbon assurance (treated group) with those that do not adopt carbon assurance (control group). We conduct a t-test to compare whether the means of the covariates significantly differ across the treated and control groups (Table 11, Panel B). The results show that any differences in all the covariates between the two groups are insignificant, suggesting that the covariates are balanced between the treated and control groups. We re-estimate the baseline regression models based on the propensity score matching estimates (Table 11, Panel A). Consistent with our main findings, we report a significant positive relationship between carbon assurance (both level and percentage) and firm-level credit ratings.

An instrumental variable method is also recommended to mitigate potential endogeneity problems (Hill et al., 2020). Following prior studies (Fu et al., 2020; Liu et al., 2015; Saeed et al., 2022), we employ the industry level of carbon assurance as the instrumental variable. The instrumental variable must be relevant (i.e., highly correlates with a firm's carbon assurance level) and exclusive (i.e., only influences the firm's credit rating through its carbon assurance level) (Ebbes et al., 2016; Jiraporn et al., 2014). We argue that the carbon assurance level of a firm is highly correlated with that of its industry peers due to their similar business nature (Liu et al., 2015). However, the industry carbon assurance level should not influence a firm's credit rating, except through the firm's carbon assurance level.

We estimate a 2SLS analysis. In the first stage, we estimate the carbon assurance level using the industry average level, controlling for all the relevant variables as in our baseline model. Results show a significantly positive coefficient for the average carbon assurance level (0.148, z-statistic = 2.10) (Table 12, model (1)). In the second stage, we regress firm credit ratings on the predicted carbon assurance level from the first stage and other control variables and also document a significantly positive relationship between the predicted carbon assurance level and firm-level credit ratings (1.388, z-statistic = 5.47) (Table 12, model (2)). The result supports our main finding, confirming a positive influence of carbon assurance on credit ratings.

Lastly, since we only include firms in our sample that report information to the CDP, which may introduce potential self-selection bias, we also employ Heckman's two-stage selection test to address this bias (Bui et al., 2021; Fan et al., 2021). Following Fan et al. (2021) and Bui et al. (2021), in the first stage, we estimate the probability that firms will participate in the CDP. In addition to the control variables in our baseline model, following the study conducted by Fan et al. (2021), we further control for: (i) the previous year's decision to report carbon information (a dummy variable that equals 1 if a firm reports carbon information to the CDP last year, and 0 otherwise—CDISC_t−1); and (ii) the industry-average carbon disclosure index (INDCDISC) in this stage. Research also shows that firms that provide carbon information to the CDP in the previous year tend to join the CDP in the current year (Fan et al., 2021). According to Fan et al. (2021), managers' disclosure decisions can be driven by industry practices, thus, we also control for the industry INDCDISC variable. Our first stage results show support for those relationships (Table 13, model (1)). In the second stage, we include the inverse Mills ratios (INVERSE_MILLS_R) obtained from the first stage in our models. The inverse Mills ratios are insignificant for both models (carbon assurance level and percentage). Importantly, the positive relationship between carbon assurance and firm credit rating still holds (Table 13, models (2) and (3)).